There’s nothing worse than a glut of inventory that you can’t unload. That has been the fear all along the PC supply chain since last year, causing OEMs, manufacturers, ODMs and resellers to be exceptionally cautious in the management of their inventories.
Now, however, the problem has been reversed. All of a sudden demand is accelerating faster than many vendors can ramp up production, causing shortages and supply constraints all along the chain.
The sector hit hardest is the GPU business, where both nVidia and ATI, AMD’s (NYSE: AMD) graphics division, are facing a severe production shortfall due to the inability of their overseas supplier to provide enough parts.
Neither nVidia (NASDAQ: NVDA) nor ATI nor make their own semiconductors. That job falls to the Taiwan Semiconductor Manufacturing Company, or TSMC. Unfortunately for both chip makers, especially ATI, TSMC is falling down on the job. It has had an extremely difficult time transitioning from the older 55 nanometer (nm) manufacturing process to the newer 40nm process.
It was the foremost issue on the mind of nVidia CEO Jen-Hsun Huang during the firm’s earnings call last week.
“We shipped everything we had in Q2. We shipped everything we had in Q3, and if our customers were willing to take chairs and furniture, I would’ve shipped that, too. There was just nothing left. And we were still short by a few million parts,” he said on the conference call with analysts following the release of excellent third-quarter numbers.
Dave Baumann, product manager for ATI’s new 40nm 5800 Radeon video chip series, is feeling a similar pinch.
“We’re not at a point where we have to hold back on the channel to fulfill system builder demand. Absolutely there is a high demand across the board from everyone. Could we be selling more if the supply was there? Yeah, absolutely,” he told InternetNews.com
“Black magic”
The process of making chips is long, slow and exacting. “It’s black magic and they have to get the formula just right,” as Jon Peddie, president of Jon Peddie Research puts it. You have no way of knowing how many working chips will come off the wafer until it’s done. Chips are “spun” onto a wafer 300 millimeters, or roughly one foot, in diameter.
It takes about a week to make the wafer, according to Peddie, and then comes the real hard part — the slicing and preparing of the chips, and the testing and validation. All told, it can take a month to make a chip, and only at the end will the manufacturer know if it actually works or goes to the trash heap.
TSMC has only recently gotten its yields to 40 percent per wafer, according to rumors around the semiconductor blogs, which means 60 percent of what it makes is junk. And that’s an improvement over earlier months.
There is a secondary option: AMD’s former fab, Globalfoundries. It was going to be a big chip maker in its own right when AMD spun off the fabrication plants earlier this year. The acquisition of Chartered Semiconductor makes it the second-largest fab behind TSMC.
But Global uses a different method of making chips, something called SoS, or Silicon on Insulator, to make CPUs. GPUs are a different design and use a manufacturing process called bulk CMOS.
The designs are inherently different, and not easy to shift from one to the next, said Peddie. “It would take at least a year for Global to set up a bulk CMOS line. They will do it, that’s known. They will have a line and when they do ATI is dumping TSMC and going to Global. But it will take time,” he said.
nVidia has said nothing about finding a new supplier. It was not asked on the earnings call and the company declined to discuss finding a new parts supplier.
Retail impact
How hard can this impact retail? It depends on how bad OEMs and customers want those screaming new chips, according to analysts.
“The alternative is that [GPU vendors] can probably sell older 55nm parts to fill demand. They’d probably have to take a hit on price and margin because OEMs won’t be happy about it but they’d rather sell something [than] nothing,” Peddie said.
The big question mark for him is inventories. “If the inventory level is still low, it will have a clear impact on the holiday season. If there’s some slack in the channel, then the channel guys can run that down and refill it in Q1. And there’s some good reason to do that, since they don’t get caught with inventory,” he explained.
Baumann said there is enough supply for now to satisfy system builders.
“I’m not convinced there’s that much of a backlog of them having to push back releases of PCs if they are seeking one of our 40nm parts,” he said. “I’ve seen the stories but I’m not convinced they are that accurate. I don’t think people have started delaying shipments of PCs because of this. We’re not at the point where the channel is completely bled dry.”
Baumann added he expects that some changes made by TSMC in recent weeks will ease inventory problems before Christmas, and there will be more products in the channel in early December.
Other shortages
The GPU shortage is the most prominent because both major players are complaining about it, but it’s not alone. Other parts are in short supply, causing price spikes. Memory on the DRAMeXchange, which follows memory prices, has spiked by as much as 60 percent in one week. There are shortages of both DDR2 and DDR3 memory.
According to DRAMeXchange, the DDR2 shortage stems from manufacturers migrating to DDR3 (or trying to) and greater-than-expected PC demand in Q3. The DDR3 market remains stable, but the price will likely rise due to increased demand and use of more memory in systems.
Other parts are in short supply as well, according to research firm iSuppli. Its electronic parts tracking service, called Component Price Tracker (CPT), rose in the third quarter as the lead times for nearly all parts categories tracked by iSuppli — like analog parts, capacitors, standard logic ICs, oscillators, resistors, most memory devices and others — increased due to rising demand.
In some cases, it’s just been dumb luck. A power disruption at Corning’s facility in Taiwan shut down LCD glass manufacturing for two days, which iSuppli predicted would reduce the availability of glass but not drive up panel prices.
But Dean McCarron, president of Mercury Research, said OEMs may shift the configurations around in the PC if they don’t get the parts they want. “A GPU is one of those discretionary spending items for an OEM. They can’t ship a PC without a processor and they can’t ship without memory but they can ship it without a discrete GPU,” he told InternetNews.com.
McCarron said this is a recurrent cycle of shortage, adjustment, surplus and crash. The surplus and crash took place last year and now we’re in the shortage stage. It won’t get any better initially, he predicted.
“We’re coming into 2010 with one of the freshest product refreshes we’ve seen across the board. Intel (NASDAQ: INTC) is going to Core i3 and i5, the graphics guys purged old stuff, and we’re moving to DDR3. So we’re going to have some very fresh platforms. There’s no risk of a shortage of computers,” he said. “What will be affected is what those PCs look like. We might see more integrated instead of add-in parts, or older parts mixed with new.”
